Bayesian Estimation and Tracking: A Practical Guide

4.1 The Overall Simulation/Estimation/Evaluation Process

Setting up software code for a specific tracking problem scenario can sometimes be a daunting task. In Figure 4.1, we present a diagram of the process blocks that must be addressed. First, one must build a scenario simulator that generates the truth data for the states to be estimated. This truth data set must then be transformed into a set of truth observations, prior to adding noise.

Figure 4.1 Methodology for development and evaluation of tracking algorithms.

By way of example, for tracking a vehicle in three Cartesian dimensions, the vehicle truth position and velocity components in Cartesian coordinates (which we call the truth track) must be generated. If the observations are from a radar, the Cartesian truth track must undergo a Cartesian-to-spherical conversion, generating the noiseless spherical truth tracks that include range, bearing, elevation, and their respective rates. Then, the specific radar observations (e.g., range, bearing, elevation, and/or Doppler range rate) can be isolated from the spherical truth. In exercising a tracking filter using data obtained from a real radar, this last simulation step can be omitted. The simulation step is primarily used to test and debug the estimation algorithm implementation and for the conduct of special studies to evaluate comparative tracking algorithm performance.